a n a l y t i c a c h i m i c a a c t a 6 1 7 ( 2 0 0 8 ) 72–84

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/aca

Headspace solid-phase microextraction–gas

chromatographic–time-of-flight mass spectrometric
methodology for geographical origin
verification of coffee

Sanja Risticevic a , Eduardo Carasek b , Janusz Pawliszyn a,∗

a Department of Chemistry, University of Waterloo, 200 University Avenue West, Waterloo, Ontario, Canada, N2L 3G1
b Department of Chemistry, Federal University of Santa Catarina, Trindade, Florianópolis, 88040-900, Santa Catarina, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: Increasing consumer awareness of food safety issues requires the development of highly
Received 7 December 2007 sophisticated techniques for the authentication of food commodities. The food products
Received in revised form targeted for falsification are either products of high commercial value or those produced
25 March 2008 in large quantities. For this reason, the present investigation is directed towards the
Accepted 1 April 2008 characterization of coffee samples according to the geographical origin. The conducted
Published on line 11 April 2008 research involves the development of a rapid headspace solid-phase microextraction (HS-
SPME)–gas chromatography–time-of-flight mass spectrometry (GC–TOFMS) method that
Keywords: is utilized for the verification of geographical origin traceability of coffee samples. As
Solid-phase microextraction opposed to the utilization of traditional univariate optimization methods, the current
Gas chromatography–time-of-flight study employs the application of multivariate experimental designs to the optimization
mass spectrometry of extraction-influencing parameters. Hence, the two-level full factorial first-order design
Multivariate experimental design aided in the identification of two influential variables: extraction time and sample tem-
Coffee aroma perature. The optimum set of conditions for the two variables was 12 min and 55 ◦ C,
Principal component analysis respectively, as directed by utilization of Doehlert matrix and response surface method-
Geographical origin ology. The high-throughput automated SPME procedure was completed by implementing
a single divinylbenzene/carboxen/polydimethylsiloxane (DVB/CAR/PDMS) 50/30 ␮m metal
fiber with excellent durability properties ensuring the completion of overall sequence of
coffee samples. The utilization of high-speed TOFMS instrument ensured the completion of
one GC–MS run of a complex coffee sample in 7.9 min and the complete list of benefits pro-
vided by ChromaTOF software including fully automated background subtraction, baseline
correction, peak find and mass spectral deconvolution algorithms was exploited during the
data evaluation procedure. The combination of the retention index (RI) system using C8 –C40
alkanes and the mass spectral library search was utilized for the confirmation of analyte
identity in the reference authentic Brazilian coffee sample. The semi-quantitative results
were then submitted to statistical evaluation, namely principal component analysis (PCA)
for the establishment of geographical origin discriminations.
© 2008 Elsevier B.V. All rights reserved.

∗
Corresponding author. Tel.: +1 519 888 4641; fax: +1 519 746 0435.
E-mail address: janusz@uwaterloo.ca (J. Pawliszyn).
0003-2670/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.aca.2008.04.009
 a n a l y t i c a c h i m i c a a c t a 6 1 7 ( 2 0 0 8 ) 72–84
1. Introduction
The globalization of food markets together with easiness asso-
ciated with import and export of food commodities have
caused increasing aspiration towards improving the current
traceability systems. According to the EU Commission, trace-
ability is defined as “the possibility to find and follow the trace,
throughout all the stages of production, processing and dis-
tribution of a foodstuff, feedstuff, an animal destined for food
production or a substance destined to be incorporated in food-
stuff or feedstuff or with a probability of being used as such”
(REGULATIONS (EC) nos. 1830/2003 & 882/2004) [1]. Thus, for
the successful utilization of traceability system, the certifi-
cation of geographical origin attributes is a must, since the
authenticity and quality parameters are often associated with
a particular geographical origin and/or production area. For
this reason, all divisions within food sector are in high demand
for the development of analytical strategies capable of tracing
back the food commodities to their production area.
Coffee is the world most popular beverage after water, con-
sidering that over 400 billion cups are consumed annually
[2]. In addition, coffee is one of the most important agricul-
tural products in the international trade, whose production is
responsible for putting into motion approximately 35 billion
US dollars on annual basis. The coffee aroma is character-
ized by the presence of a wide range of volatiles belonging
to several classes of compounds such as furans, pyrazines,
ketones, alcohols, aldehydes, esters, pyrroles, thiophenes, sul-
phur compounds, benzenic compounds, phenols, pyridines,
thiazoles, oxazoles, lactones, alkanes, alkenes, and acids [3,4].
These components are present in variable concentrations and
each of them contributes uniquely to the final aroma quality.
The research dedicated to the recognition of coffee adul-
teration practices has been conducted for years. One of the
most commonly utilized strategies in this field is the possibil-
ity to discriminate coffees based on the coffee bean variety.
Even though, at least 66 species of the genus Coffea L. have
been identified so far, only two varieties with pronounced
economical and commercial importance are extensively culti-
vated in coffee producing regions: C. arabica L. (arabica coffee)
and C. canephora Pierre (robusta coffee) [5]. Arabica coffee
beans are generally characterized by more pronounced qual-
ity and consumer acceptance, consequently they command
higher commercial value prices and are therefore subject
to fraudulent practices [4,6]. The characterization of coffee
samples according to the bean variety has been established
by several authors who accomplished this task either by
focusing on the metal content [6,7] or volatile/semivolatile
compound compositions [4,8–11]. In addition to utilizing
the qualitative comparison procedure based on the com-
positions of volatile/semivolatile and metal contents and
submitting the semi-quantitative results to statistical anal-
ysis, several authors performed the differentiation studies
based on the determination of characteristic marker com-
ponents that are predominant in one of the two species of
interest. Accordingly, (−)-2-methylisoborneol was identified as
a key aroma compound in robusta coffee [12] and trigonelline
(N-methylnicotinic acid), nicotinic acid and caffeine were also
assessed in terms of representing the marker compounds
73
responsible for the differentiation between the two varieties
[13].
In addition to discrimination based on coffee bean variety
as one of the methods for confirmation of coffee authentic-
ity, another important approach that can be used to meet this
objective is geographical origin declaration. The major pro-
ducers of coffee beans are South American countries such as
Brazil and Colombia, however coffee beans are produced to a
large extent in Central America, Africa and Asia. The princi-
pal consumers of coffee beans are on the other hand European
and North American countries, where at this point it is impor-
tant to emphasize the significance of quality certification
procedures based on patterns and historical specifications
for both the internal and external coffee markets. Consider-
ing that in recent years the increasing practices associated
with selling coffees on the basis of their geographical origin
and falsifying the product declaration have been detected,
the need for analytical methodologies capable of verifying
the geographical origin of coffee is therefore outstanding. In
order to meet this demand, several authors’ publications were
directed towards meeting this objective [9,11,14,15], however,
the relatively small number of publications in this area still
requires the development of highly sophisticated and modern
techniques capable of ensuring the geographical origin coffee
authenticity.
Gas chromatography–mass spectrometry is an ideally suit-
able technique for the analysis of naturally volatile and
semivolatile analytes in food commodities, including coffee
[4,16]. In addition, gas chromatographic data is easily employ-
able into multivariate analysis softwares, such as principal
component analysis (PCA), linear discriminant analysis (LDA)
and partial least squares (PLS). While the non-target deter-
mination of aroma-contributing coffee constituents has been
traditionally achieved through implementation of gas chro-
matographic methods, recently increasing the throughput of
analytical procedure has become the major objective of any
GC method development. In the current study, this objec-
tive was achieved by utilizing narrow-bore column fast GC
in combination with high-speed TOFMS mass analyzer. To
the best of the authors’ knowledge, no previously published
documents focusing on the determination of volatile and
semivolatile coffee constituents and subsequent assessment
of coffee authenticity using high-speed GC-TOFMS system are
available in literature.
As was already elaborated previously in this document,
coffee aroma is attributed to a vast number of compounds
present at variable concentrations and the list of identified
compounds in such a complex matrix is far from completion.
For this reason, the analyte preconcentration prior to chro-
matographic separation is absolutely necessary. Distillation
techniques (steam and vacuum) have been widely used in the
analysis of coffee samples [17,18]. In addition, purge and trap,
static headspace, headspace sorptive extraction and stir bar
sorptive extraction sample preparation techniques have also
been utilized as extraction procedures prior to gas chromato-
graphic analysis [19–21].
Solid-phase microextraction (SPME) is a fairly recent
sample preparation technique developed by Pawliszyn and co-
workers in 1990 [22]. The development of SPME has addressed
the need for a rapid sample preparation technique, as the fol-
74 a n a l y t i c a c h i m i c a a c t a 6 1 7 ( 2 0 0 8 ) 72–84
lowing procedures are integrated into a single solvent-free
step: (i) sampling; (ii) extraction; (iii) concentration and (iv)
sample introduction. For this reason, the number of SPME
applications in food analysis is vast, and recently SPME has
been successfully utilized for the isolation of coffee aroma
compounds [3,4,9–11,15,21,23–32].
When utilizing SPME, the SPME method development pro-
cedure needs to be conducted efficiently as the effectiveness
of analyte preconcentration depends on many parameters,
such as: type of fiber coating, extraction mode, extrac-
tion conditions, sample volume and desorption conditions
[33]. Traditionally, the SPME method optimization has been
conducted by considering one parameter at a time and keep-
ing all other variables constant, the method referred to as
univariate optimization design [34]. However, recently mul-
tivariate optimization design methods combining first-order
(to detect influential variables) and second-order (to opti-
mize influential variables) experimental designs have become
increasingly popular as they allow the simultaneous study of
several control variables [35]. Most commonly, the multivari-
ate optimization design is carried out by utilizing two-level
full factorial and Doehlert matrix design accompanied by
response surface methodology, for first-order and second-
order designs, respectively. This approach was employed
in the current study as well and to the best of authors’
knowledge, there are no published literature findings on the
utilization of this combination of multivariate optimization
designs for SPME method development associated with the
extraction of volatile and semivolatile coffee aroma con-
stituents.
The objective of the current research study is dedicated
towards the development of high-throughput automated HS-
SPME–GC–TOFMS methodology to allow qualitative screening
study of volatiles and semivolatiles found in coffee. The SPME
method development is to be established by utilizing multi-
variate optimization design methods: two-level full factorial
and Doehlert matrix design accompanied by response surface
methodology. The optimized method will be then applied to
the analysis of real coffee samples collected from both pro-
duction and non-production regions, after which the acquired
data will be submitted to PCA statistical evaluation to establish
the geographical origin traceability and authenticity of coffee
samples.
2. Experimental
2.1. Samples
The Arabica coffee samples considered for geographical ori-
gin discrimination were collected from: (i) producing regions
such as Brazil and Colombia and (ii) Canada (where the
geographical specifications were acquired from coffee dis-
tributing markets). The total number of samples submitted
to geographical origin classification is summarized in the
following way: (i) production area, Brazil – 11 samples; (ii)
production area, Colombia – 8 samples; (iii) Canada collec-
tion comprised of South American samples (Brazil – 2 samples
and Colombia – 12 samples), Central American samples (Costa
Rica – 3 samples and Guatemala – 4 samples), African sam-
ples (Ethiopia – 3 samples) and Asian samples (Indonesia – 4
samples).
2.2. Analytical reagents and supplies
The alkane mixtures containing C8 –C20 and C21 –C40 straight-
chain alkanes of 40 mg L−1 concentration in hexane and
toluene, respectively, were purchased from Fluka (Buchs,
Switzerland). Helium of purity 5.0 (Praxair Canada, Mis-
sissauga, Ontario, Canada) was utilized as the GC carrier
gas. The SPME fiber optimization step within SPME
method development was carried out by testing com-
mercially available silica SPME fibers obtained from Supelco
(Bellefonte, PA, USA) and coated with the following poly-
mers: polydimethylsiloxane (PDMS, 100 ␮m and PDMS,
30 ␮m), polydimethylsiloxane/divinylbenzene (PDMS/DVB,
65 ␮m), divinylbenzene/carboxen/polydimethylsiloxane
(DVB/CAR/PDMS, 50/30 ␮m), carboxen/polydimethylsiloxane
(CAR/PDMS, 75 ␮m), polyacrylate (PA, 85 ␮m), and car-
bowax/divinylbenzene (CW/DVB, 70 ␮m). All fibers were
conditioned according to the manufacturer’s recommenda-
tions prior to their first use. Clear glass crimp cap 10 mL SPME
vials (22 mm × 46 mm) and caps equipped with polytetraflu-
oroethylene (PTFE)/silicone septa (20 mm) were purchased
from MicroLiter Analytical Supplies, Inc. (Suwanee, GA, USA)
and Canadian Life Science (Peterborough, Ontario, Canada),
respectively.
2.3. Equipment
A CombiPAL SPME autosampler (Leap Technologies, Carrboro,
NS, USA) equipped with an agitator and fiber conditioning
station was utilized to allow completely automated SPME
procedure. The autosampler was employed in combination
with a gas chromatograph 6890 (Agilent Technologies, Palo
Alto, CA, USA) coupled to a Pegasus III mass spectrom-
eter (LECO, St. Joseph, MI, USA). The mass spectrometer
was equipped with a time-of-flight (high-speed data acqui-
sition rate option) mass analyzer. The separation of volatile
and semivolatile compounds was achieved by utilizing the
5% diphenyl/95% dimethylpolysiloxane SLB-5 column (10 m
length, internal diameter 180 ␮m, 0.18 ␮m film thickness) pur-
chased from Supelco (Bellefonte, PA, USA). A high-pressure
Merlin Microseal septumless injection kit (Merlin Instrument
Co., Half Moon Bay, CA, USA) was obtained from Chromato-
graphic Specialties (Canada). After the completion of SPME
fiber optimization experiments, the 23-G needle size super
elastic SPME fiber coated with best suitable polymer was used
in combination with this septumless injector for the comple-
tion of overall sequence of coffee samples.
2.4. Experimental conditions
The fiber optimization experiment was accomplished by test-
ing the efficiency of commercially available fiber coatings
listed in Section 2.2. For this particular optimization experi-
ment, the samples were prepared by transferring 1-g portions
of ground coffee (Van Houtte, 100% Colombian dark roast
coffee) obtained in grocery supermarket to 10-mL vials. The
incubation and extraction procedures were completed by uti-
 a n a l y t i c a c h i m i c a a c t a 6 1 7 ( 2 0 0 8 ) 72–84
lizing sample temperature and agitation speed of 60 ◦ C and
500 rpm, respectively. The employed incubation and extrac-
tion times were 10 min and 20 min, respectively, whereas the
thermal desorption of analytes in the GC injector port was
enabled for 1.5 min at 250 ◦ C. After each extraction-desorption
cycle, the fiber was cleaned in the fiber conditioning station for
5 min.
Both fiber selection and multivariate experimental design
optimization experiments were carried out by employing the
same GC and MS conditions. The splitless mode injections
were carried out at constant injector temperature of 250 ◦ C.
The carrier gas, helium was used at a constant flow rate of
1.5 mL min−1 . The oven temperature program applied in the
analysis consisted of utilizing 40 ◦ C initial temperature for
1.5 min, after which the temperature was raised at 40 ◦ C min−1
rate to 295 ◦ C, for a total GC run time of 7.9 min. The trans-
fer line temperature was held at 295 ◦ C. The column eluent
was submitted to electron impact (EI) ionization, while the ion
source temperature was kept at 220 ◦ C. The detector voltage
was 1700 V. The data acquisition rate was 10 spectra s−1 and
mass fragments were collected in m/z 35–450 range.
2.5. Autosampler, data acquisition and statistical
analysis softwares
The CombiPAL autosampler was operated using the PAL Cycle
Composer with Macro Editor software (version 1.4.0.). After
the collection of GC–MS data, the total ion current (TIC)
chromatograms were processed using the LECO ChromaTOF
automated data processing software (version 2.32). The com-
pounds considered for method optimization were identified
by comparing their mass spectra to the mass spectra of ref-
erence compounds in the US National Institute of Standards
and Technology (NIST), Terpene, and Flavor libraries (the latter
two were included in the ChromaTOF software). With regards
to the compounds identified in actual coffee samples sub-
mitted to differentiation study, the identity confirmation was
based on: (i) the mass spectral comparison described above
and (ii) the calculation of experimental retention indices and
their comparison with the literature retention index data. The
multivariate optimization design step composed of: (i) two-
level full factorial and (ii) Doehlert matrix which was utilized
for SPME method development was completed by employ-
ing Statistica software. The characterization of coffee samples
according to geographical specifications was established using
PCA evaluation included within SPSS software for Windows,
version 15.0.
3. Results and discussion
3.1. Fiber selection experiment
The sensitivity of SPME extraction technique depends greatly
on the value of the distribution constant of analytes parti-
tioned between the sample and fiber coating material [33].
For this reason, the extraction efficiency of all commercially
available SPME fiber coatings described in Section 2.2 was
tested such that the coating having highest affinity towards
volatile and semivolatile coffee constituents can be selected
75
Fig. 1 – Performance characteristics of CAR/PDMS and
DVB/CAR/PDMS fibers for isolation of volatile and
semivolatile constituents. The peak numbers are
associated with compounds presented in Table 3.
for the analysis of real coffee samples. For this purpose, 15
analytes (see corresponding peak numbers 3, 7, 9, 13, 15, 17,
23, 36, 43, 68, 83-84, 89, 93 and 102 in the list of tentatively
identified compounds illustrated in Table 3) being character-
ized by different volatilities and polarities were selected across
the GC chromatogram and the total sum of their peak areas
was used to identify the SPME coating having best perfor-
mance characteristics. The results illustrated that CAR/PDMS
and DVB/CAR/PDMS fibers were most suitable for this study,
as the total sum of the peak areas was highest for these
two fibers. These two fiber coatings are commonly used in
flavour analysis [23] and for this reason, the more detailed
comparison of these two fibers was conducted and results
illustrated in Fig. 1, leading to conclusion that CAR/PDMS fiber
assembly exhibits better sensitivity for low molecular weight
analytes. On the other hand, the triple phase DVB/CAR/PDMS
coating material proved to have better performance charac-
teristics for sufficient isolation of analytes having wide range
of physico-chemical properties. The similar outcomes were
concluded by Mondello et al. [3] and Akiyama et al. [24] in
their studies associated with the optimization of SPME condi-
tions for isolation of coffee aroma constituents. Based on the
data evaluation completed within this particular optimization
experiment, DVB/CAR/PDMS fiber was utilized in all further
optimization/real analysis experiments.
3.2. Multivariate experimental design
The traditionally utilized univariate optimization design
involves the performance of “one factor at a time” experi-
ments, thus offering the possibility to examine the effect of
one variable at a time since all other variables are kept con-
stant during a particular optimization experiment, except the
one being studied. For this reason, this traditional method
might overlook the possibility of potential variable interac-
tion, the issue which arises when the variables kept constant
in previously performed optimization experiments change
[34]. Consequently, after the completion of “univariate” SPME
76 a n a l y t i c a c h i m i c a a c t a 6 1 7 ( 2 0 0 8 ) 72–84
method development, some SPME parameters might require
re-examination followed by repetition of corresponding opti-
mization experiments.
Multivariate designs on the other hand, offer the simul-
taneous variation of several control variables, consequently
reducing the number of experiments to be performed in SPME
optimization procedure. In addition, multivariate designs
offer the possibility of distinguishing those interactions
among variables that would not be detectable by classical
experimental design [36].
The effectiveness of analyte preconcentration using the
SPME technique depends on the following parameters: (i)
amount of sample; (ii) incubation time; (iii) extraction time;
(iv) incubation/extraction temperature and (v) agitation speed
[33]. For this reason, the extent to which each of these vari-
ables affects the efficiency of SPME procedure was examined
via application of multivariate optimization design.
3.2.1. Two-level full factorial first-order design
The first step of the optimization procedure involved the
utilization of two-level full factorial design to assess the
importance of each variable (from the list illustrated in Sec-
tion 3.2.) on the SPME extraction efficiency. Consequently, 25
full factorial design was applied to identify the most important
variables and potential variable interactions.
Each of the five studied variables was examined at two lev-
els (minimum and maximum ones), for which the values were
arbitrarily selected to cover the wide range of experimental
conditions. 25 factorial design required the performance of 32
experiments which were done in duplicates for a total num-
ber of 64 experiments. Based on the selection of minimum
and maximum levels, the variables were also examined in the
corresponding central points of a factorial design. The cen-
tral point experiments were performed in four replications,
leading to the overall number of 68 experiments required for
this particular optimization step. Table 1 illustrates the mini-
mum, maximum and central point ranges for each variable of
interest.
The data evaluation pertaining to this particular exper-
iment was accomplished by integrating and evaluating the
peak areas corresponding to 30 compounds having different
retention times and being characterized by various polar-
ity and volatility characteristics which were selected across
the entire GC chromatogram. This is a common approach
being employed when developing methods to allow non-target
analysis of complex food matrices such as coffee in this par-
ticular case. In addition, this technique considers reasonable
assumption that either a polar or a non-polar, volatile or
Table 1 – The minimum, maximum and central point ranges selected for each variable of interest
Variables
Minimum
Amount of coffee sample (g)
Incubation time (min)
Extraction time (min)
Incubation/extraction temperature (◦ C)
Agitation speed (rpm)
Fig. 2 – Pareto chart of standardized effects of 25 factorial
design for total chromatographic peak area corresponding
to 30 analytes selected for SPME method optimization.
semivolatile compound could be the marker for establishing
the geographical origin discrimination of food samples [37].
The response based on the sum of the peak areas is one of
the most useful parameters for the optimization of SPME con-
ditions, and hence it was considered as the end point in this
study as the sum of peak areas corresponding to 30 analytes
selected for method optimization. An analysis of variance
(ANOVA) was performed to assess whether the experimental
variables examined in this study were affecting the perfor-
mance of HS-SPME procedure in a significant way. The variable
effects and interactions are summarized in the Pareto chart
illustrated in Fig. 2.
As demonstrated in Fig. 2, extraction time and sample
temperature are statistically significant at the 95% confidence
level. Furthermore, the results summarized in the Pareto chart
indicate that increasing these two factors will increase the
analytical signal as well. The positive effect of extraction time
increase can be attributed to the fact that the equilibrium is
not reached even after 30 min for less volatile compounds eval-
uated in this study. With regards to the sample temperature,
two opposite phenomena occur upon increasing sample tem-
perature. Higher extraction temperature enhances the release
of analytes from sample matrix into the headspace, hence the
rate of analyte transfer towards the fiber is increased [37]. On
the other hand, by increasing the sample temperature, the
distribution constant of the analyte between the headspace
and fiber coating decreases, hence decreasing method sensi-
Levels
Central point Maximum
0.5 1.0 1.5
0.5 10.25 20
0.5 15.25 30
40 60 80
0 250 500
 a n a l y t i c a c h i m i c a a c t a 6 1 7 ( 2 0 0 8 ) 72–84
Fig. 3 – Pareto chart of standardized effects of 25 factorial
design for total chromatographic peak area corresponding
to 15 more volatile analytes selected for SPME method
optimization.
tivity [38,39]. The Pareto chart from Fig. 2 illustrates that upon
increasing the sample temperature, the effect of more promi-
nent analyte transfer towards the fiber is more significant than
the one associated with the decrease of distribution constant.
The effect of sample temperature was consequently eval-
uated in more detail by considering the sum of peak areas of
first 15 more volatile compounds as the response for Pareto
chart illustrated in Fig. 3. This finding is contrary to the
one presented in Fig. 2, as the higher significance of sam-
ple temperature as compared to extraction time is observed.
As expected, the extraction time was not identified as most
influential variable this time due to the shorter equilibration
times of more volatile analytes. Furthermore, increasing sam-
ple temperature has negative effect on the SPME extraction
efficiency in this case, which is in accordance with the fact that
the effect of decreasing distribution constant is more critical
for lighter compounds.
3.2.2. Doehlert design
The results accomplished by the implementation of full facto-
rial design demonstrate that at the 95% confidence level, the
sample amount, incubation time and agitation speed are not
effecting the SPME extraction efficiency to a significant extent.
For this reason, the values associated with these three param-
eters were fixed in accordance to previously selected ranges
in Table 1 to 1 g, 5 min and 500 rpm, respectively. The highly
influencing variables, extraction time and sample tempera-
ture needed to be further optimized through the utilization of
second-order Doehlert design.
The two-variable Doehlert matrix was composed to allow
the five-level and three-level examinations of extraction time
and sample temperature, respectively. The experimental fields
composing the Doehlert matrix are summarized in Table 2.
Seven experiments were required to be performed in dupli-
cates for this particular two-variable matrix having the shape
of regular hexagon with a central point.
The sum of the peak areas corresponding to same 30 pre-
selected analytes in combination with experimental fields
77
Table 2 – Experimental fields for two-variable Doehlert
matrix
Experimental variables
Incubation/extraction Extraction time
temperature (◦ C) (min)
1 40 5
2 40 13
3 55 9
4 55 1
5 55 17
6 70 13
7 70 5
presented in Table 2 were used to obtain the response sur-
face plots. The response surface plot associated with the sum
of peak areas corresponding to both more volatile and less
volatile compounds is illustrated in Fig. 4. From the illustrated
results, it can be concluded that the optimum set of conditions
for all evaluated analytes consists of 12 min and 65 ◦ C. The
effect of these two variables was further investigated by build-
ing a response surface plot (Fig. 5) for 15 more volatile analytes,
exclusively. The optimum value for extraction time parame-
ter as directed by the results illustrated in response surface
plot, Fig. 5 is 12 min in this case as well. However, the effect
of sample temperature on the extraction efficiency is quite
different than when considering all analytes, since the opti-
mum temperature for the extraction of more volatile analytes
is lower than 40 ◦ C. Accordingly, the results observed through
full factorial design are confirmed by utilizing response sur-
face methodology as well, since it is clear that the optimal
temperature condition is affected by analyte volatility. Since
the objective of this study is to perform non-target screening
of coffee constituents across the entire GC chromatogram, the
sample temperature of 55 ◦ C was selected as a compromise
condition to ensure the efficient extraction of both the more
volatile and less volatile analytes.
Fig. 4 – Response surface associated with both more volatile
and less volatile analytes selected for method optimization.
78 a n a l y t i c a c h i m i c a a c t a 6 1 7 ( 2 0 0 8 ) 72–84
Fig. 5 – Response surface associated with 15 more volatile
analytes selected for method optimization.
3.3. Analysis of volatile and semivolatile coffee
constituents
After the completion of method development section, the
optimized HS-SPME–GC–TOFMS method was applied to the
analysis of real coffee samples (described in Section 2.1 of this
document) that are to be submitted to statistical geographi-
cal origin discrimination study. The utilization of high-speed
TOFMS instrument ensured the completion of one GC–MS
run of a complex coffee sample in 7.9 min. For this reason,
the SPME extraction and preconcentration procedures were
the only time-limiting steps of the coffee analysis. Additional
sophisticated features of Pegasus III high-speed TOFMS instru-
ment were thoroughly summarized by Setkova et al. in the
project directed towards the classification of ice wine sam-
ples according to various quality-influencing attributes such
as grape variety and geographical origin [37,40]. Briefly, the
large data set acquired in the previous ice wine characteri-
zation study was easily processed by utilizing the benefits of
fully automated background subtraction, baseline correction,
peak find and mass spectral deconvolution algorithms. The
ChromaTOF software provided automatic assignment of most
unique and least coeluting quantification mass to a particular
peak. Furthermore, the “compare to reference” feature incor-
porated within ChromaTOF software allowed the comparison
of the particular sample undergoing processing procedure to
the pre-selected reference sample, such that the comparison
results are sorted in the following fashion: “match”, “out of
tolerance”, “contaminant” and “not found” [37]. This compar-
ison strategy was cost-effective and beneficial also for the
current non-target screening study whose objective is the
identification of suitable markers for geographical origin cof-
fee traceability.
Having this in mind, the coffee sample originating from
Brazilian authentic production region having rich volatile and
semivolatile chromatographic profile was selected as the ref-
erence sample, to which all the other coffee samples involved
in characterization study were compared. The tentative com-
pound identification was performed by utilizing the following
two procedures: (i) the comparison between experimental
and reference/library mass spectra and (ii) the comparison
between experimental and literature retention indices. As it
has been well documented in previously published works, the
collection of GC retention indices represents an extremely use-
ful tool for the confirmation of analytes’ identity in complex
food matrices [41]. These indices are independent from all the
experimental conditions, except the polarity of the GC sta-
tionary phase utilized in a particular study. For this reason,
their utilization is fundamental for interlaboratory compari-
son. When dealing with complex aroma of food and flavour
matrices, it is quite common for a number of compounds
to exhibit similar fragmentation patterns, in which case, the
identification of compounds based on retention index system
is crucial for the analyte identity conformation [42].
By utilizing the two identification methods specified here,
the coffee aroma associated with authentic reference sample
from Brazil is described by 102 volatile and semivolatile ana-
lytes with specifications illustrated in Table 3. The pyrazine
fraction which is formed during the roasting procedure trans-
formation of green coffee beans into the roasted ones is
described by 25 positively identified compounds. In addition,
the coffee aroma associated with this reference sample can be
attributed to extensive ketone, aldehyde, furan, pyridine and
pyrrole fractions.
3.4. Statistical evaluation of data and geographical
origin discrimination of coffee samples
Prior to utilizing the statistical evaluation tools for verification
of geographical origin attributes, 29 coffee aroma-contributing
compounds (see the underlined compounds in Table 3) were
evaluated in all samples that are to be submitted to classifi-
cation study by performing automatic integration procedure.
In addition, the automatic integration and peak assignment
steps were inspected leading to the necessity to perform corre-
sponding manual procedures in some cases. The average peak
areas were calculated based on three replicates for which the
analysis of real samples was performed. In accordance with
previously published study by Zambonin et al. [9], the data
were re-processed by employing “internal normalization” pro-
cedure such that the areas of each of the underlined 29 peaks
in Table 3 were expressed as a percentage of the sum of their
total average peak areas. PCA statistical processing was then
performed to study the main sources of variability between
the coffee samples originating from various cultivation areas
and detect the potential relationships/variables responsible
for differentiation.
3.4.1. Authentic sample collections
As already stated in Section 1 of this document, the major pro-
ducers of coffee beans are South American countries such as
Brazil and Colombia. For this reason, the preliminary objective
associated with geographical origin characterization of coffee
samples was directed towards submitting the normalized data
corresponding to samples received from authentic/production
regions to statistical analysis. Overall, the authentic Brazil-
ian and Colombian collections were comprised of 11 and 8
samples, respectively.
 a n a l y t i c a c h i m i c a a c t a 6 1 7 ( 2 0 0 8 ) 72–84 79

Table 3 – Volatile and semivolatile compounds identified in the reference coffee sample
Peak no. Compound name R.T. (s) Unique mass CAS no. RI (experiment) RI (literature)

1 Acetaldehyde 10.531 42 75-07-0 477 381b

2 2-Propanone 11.631 55 67-64-1 528 471b
3 2,3-Butanedione 14.230 86 431-03-8 599 584a
4 2-Methylfuran 14.730 62 534-22-5 608 606a
5 1,2-Dimethylhydrazine 17.030 80 540-73-8 645 na
6 3-Methylbutanal 18.030 44 590-86-3 658 654a
7 2-Methylbutanal 18.830 57 96-17-3 667 645b
8 Acetic acid 20.230 75 64-19-7 682 na
9 2,3-Pentanedione 21.830 100 600-14-6 697 702a
10 2,5-Dimethylfuran 22.630 94 625-86-5 704 703a
11 3-Hydroxy-2-butanone 24.830 42 513-86-0 721 681b
12 2-Vinylfuran 25.130 39 1487-18-9 723 na
13 1-Methylpyrrole 27.730 114 96-54-8 740 731b
14 Pyrazine 28.430 96 290-37-9 744 729a
15 Pyridine 31.730 75 110-86-1 762 726b
16 Pyrrole 33.030 40 109-97-7 768 739b
17 2,3-Hexanedione 37.530 81 3848-24-6 788 781a
18 2-Methyl-5-ethylfuran 40.329 110 1703-52-2 799 791b
19 Hexanal 41.229 72 66-25-1 802 800a
20 3,4-Hexanedione 41.729 70 4437-51-8 804 773b
21 2-Methyltetrahydrofuran-3-one 44.229 55 3188-00-9 812 806a
22 4-Methylthiazole 47.829 71 693-95-8 823 791b
23 2-Methylpyrazine 51.229 37 109-08-0 833 820a
24 3-Furaldehyde 54.529 91 498-60-2 842 862a
25 Furfural 55.729 112 98-01-1 845 852a
26 2-Allylfuran 60.429 39 75135-41-0 856 na
27 2,5-Dimethylpyrrole 65.728 42 625-84-3 867 na
28 3-Methylpyridine 69.328 54 108-99-6 874 871a
29 Acetol acetate 73.428 122 592-20-1 882 na
30 Furfuryl alcohol 88.128 67 98-00-0 906 860a
31 Furfuryl formate 96.627 39 13493-97-5 875 na
32 2,6-Dimethylpyrazine 99.627 74 108-50-9 885 886b
33 2,5-Dimethylpyrazine 100.127 107 123-32-0 886 912a
34 2-Acetylfuran 100.727 96 1192-62-7 888 913a
35 2-Ethylpyrazine 104.327 96 13925-00-3 900 914a
36 2,3-Dimethylpyrazine 104.927 67 5910-89-4 902 916a
37 Gamma-butyrolactone 105.527 43 96-48-0 904 941a
38 2(5H)-furanone 111.527 84 497-23-4 924 na
39 2-Ethylpyrrole 113.526 85 1551-06-0 930 941a
40 2,5-Hexanedione 113.926 42 110-13-4 932 na
41 Benzaldehyde 122.826 42 100-52-7 961 960a
42 3-Ethylpyridine 123.426 57 536-78-7 963 955a
43 5-Methylfurfural 126.526 104 620-02-0 973 960a
44 4-Methyl-3-pentanone 129.226 142 565-69-5 982 na
45 1-Acetoxy-2-butanone 130.126 86 1575-57-1 985 na
46 3-Ethyl-2,4-dimethylpyrrole 131.226 122 517-22-6 988 na
47 2-Pentylfuran 131.826 81 3777-69-3 990 991a
48 Furfuryl acetate 136.226 91 623-17-6 1004 996a
49 2-Ethyl-6-methylpyrazine 139.125 86 13925-03-6 1014 1000a
50 2-Ethyl-3-methylpyrazine 139.625 102 15707-23-0 1015 1001a
51 Trimethylpyrazine 140.525 43 14667-55-1 1018 1002a
52 2-Formyl-1-methylpyrrole 141.225 108 1192-58-1 1020 na
53 2-Propionylfuran 141.925 95 3194-15-8 1022 988b
54 2-Butyltetrahydrofuran 143.225 39 1004-29-1 1026 na
55 2-Ethenyl-6- 144.225 53 13925-09-2 1030 na
methylpyrazine
56 2-Acetylpyrazine 146.325 108 22047-25-2 1036 1023a
57 N-Acetyl-4(H)-pyridine 146.825 124 67402-83-9 1038 na
58 3,6-Heptanedione 147.825 101 1703-51-1 1041 na
59 2-Acetylpyridine 149.125 39 1122-62-9 1045 1032a
60 2,2 -Bifuran 149.625 105 5905-00-0 1047 na
61 3,3,5-Trimethylcyclohexene 150.425 98 503-45-7 1049 na
80 a n a l y t i c a c h i m i c a a c t a 6 1 7 ( 2 0 0 8 ) 72–84

Table 3 (Continued )
Peak no. Compound name R.T. (s) Unique mass CAS no. RI (experiment) RI (literature)

62 2-Formylpyrrole 153.825 87 1003-29-8 1060 1043a

63 2,3,4-Trimethyl-2- 157.325 65 28790-86-5 1071 na
cyclopenten-1-one
64 2,6-Diethylpyrazine 161.725 145 13067-27-1 1085 na
65 2,3-Diethylpyrazine 162.824 42 15707-24-1 1088 1080a
66 2-Furfurylfuran 163.324 59 1197-40-6 1090 na
67 3-Ethyl-2,5-dimethylpyrazine 164.224 56 13360-65-1 1093 1079a
68 2-Ethyl-3,5-dimethylpyrazine 164.724 132 13925-07-0 1094 1083a
69 Furfuryl propanoate 165.224 154 623-19-8 1096 na
70 2-Acetylpyrrole 165.924 141 1072-83-9 1098 1074a
71 2-Methoxyphenol 167.024 91 90-05-1 1102 1104a
72 2-Methyl-6-[(1E)-1- 168.324 38 18217-81-7 1107 na
propenyl]pyrazine
73 2-Acetyl-3-methylpyrazine 173.524 66 23787-80-6 1128 1082a
74 Maltol 176.424 87 118-71-8 1139 1108a
75 2-Isobutyl-3- 177.724 66 13925-06-9 1144 1134a
methylpyrazine
76 5H-5-methyl-6,7- 178.924 123 23747-48-0 1149 1139a
dihydrocyclopentapyrazine
77 2-Methyl-3,5- 182.524 134 18138-05-1 1163 na
diethylpyrazine
78 2-Formyl-4,5- 183.824 105 53700-95-1 1168 na
dimethylpyrrole
79 2-Allyl-6-methylpyrazine 184.424 153 55138-64-2 1170 na
80 2-Acetyl-3-ethylpyrazine 184.924 91 32974-92-8 1172 1158a
81 2-Heptylfuran 186.124 166 3777-71-7 1177 1193a
82 Spiro[4.5]decan-2-one 186.924 168 3643-16-1 1180 na
83 2-Furfuryl-5-methylfuran 187.823 115 13678-51-8 1184 na
84 1-Furfurylpyrrole 189.023 39 1438-94-4 1189 1183a
85 Ethyl maltol 190.223 65 4940-11-8 1193 1199a
86 1-Methylpyrrolo[1,2- 195.023 190 64608-59-9 1214 na
a]pyrazine
87 Furfuryl isovalerate 198.123 69 13678-60-9 1228 1222a
88 2-Isopentyl-6- 204.623 121 91010-41-2 1257 na
methylpyrazine
89 4-Ethyl-2-methoxyphenol 210.923 79 2785-89-9 1285 1275a
90 2-[(2- 215.422 83 4437-22-3 1305 na
Furylmethoxy)methyl]furan
91 Indole 217.622 108 120-72-9 1316 1297a
92 2,6-Dimethyl-3(2-methyl-1- 218.122 122 56617-70-0 1318 na
butyl)pyrazine
93 2-Methoxy-4-vinylphenol 219.422 53 7786-61-0 1325 1309a
94 Carvacrol 219.722 146 499-75-2 1326 1317a
95 Eugenol 229.622 178 97-53-0 1374 1366a
96 ␣-Copaene 230.122 204 3856-25-5 1377 1375a
97 ␤-Damascenone 231.622 70 23696-85-7 1384 1379a
98 Difurfuryl sulfide 248.221 81 13678-67-6 1468 1463a
99 (+)-␦-Cadinene 257.521 161 483-76-1 1517 1518a
100 cis-Calamenene 258.221 159 483-77-2 1521 1525a
101 ␣-Calacorene 261.921 157 na 1541 1544a
102 Caffeine 327.118 194 58-08-2 1936 1837a

The literature retention indices displayed in the table and confirmatory retention indices were obtained from literature sources [40,43–69].
a
Retention indices from literature sources when HP-5 stationary phase utilized.
b
Retention indices from literature sources when HP-1 stationary phase utilized.

The final PCA analysis led to the extraction of two princi-
pal components (PCs) having the initial eigenvalues >1 which
contributed to 80.239% of the total variance of the data set.
The first principal component (PC1) identified as a linear
combination of 5-methylfurfural, 2,3-dimethylpyrazine and 2-
methylpyrazine accounted for 52.718% of the variance. The
second principal component (PC2) was mainly characterized
by variable 2,3-diethylpyrazine, and this extracted factor con-
tributed to 27.521% of the variance. The scatter plot of PCA
scores corresponding to this discrimination is illustrated in
Fig. 6. The conclusion that can be drawn from the illustrated
classification study is that the two authentic coffee sample
collections originating from Brazil and Colombia can be distin-
guished according to the geographical origin specifications. As
 a n a l y t i c a c h i m i c a a c t a 6 1 7 ( 2 0 0 8 ) 72–84
Fig. 6 – Scatter plot (PC1 and PC2 plotted on x and y axis,
respectively) of PCA scores associated with differentiation
of authentic Brazilian (sample codes 1–11) and Colombian
(sample codes 12–19) coffee samples.
opposed to the Brazilian samples which are very well grouped
together, the samples originating from Colombian produc-
tion regions are more scattered. At this point however, it
should be noted that in addition to geographical origin, several
other parameters were identified to influence the chemical
composition of coffee such as: (i) soil and climate condi-
tions; (ii) coffee bean processing methodology and (iii) length
and temperature of roasting process [4,27]. Unfortunately, the
standardization of such conditions is not easily established for
samples originating from various production areas and was
not achievable in the current study either. The less promi-
nent grouping of Colombian samples can be attributed to any
of the aforementioned factors. However, the outcome of the
current study illustrates that the two authentic sample collec-
tions from Brazil and Colombia can be characterized based on
geographical origin attributes despite the potential influences
and variation causes initiated by mentioned un-standardized
factors.
3.4.2. Comparison between authentic and imported
sample collections
The reliable authenticity verification of imported food com-
modities including coffee requires the implementation of
quality certification procedures capable of tracing the prod-
uct back to its production area. Furthermore, the assessment
of authenticity within the imported food sector requires the
identification of potential changes a particular food commod-
ity undergoes, especially if these changes are associated with
nutritional, flavour and quality deteriorations during stor-
age/transportation. For this reason, the objective within this
particular characterization study was to compare the volatile
coffee aroma profiles of authentic and non-authentic Brazilian
and Colombian coffee samples.
The results from finalized PCA evaluation indicated the
extraction of two PCs representing 81.170% of the total vari-
ance of the data set. PC1 represented 47.787% of the variance
and was strongly characterized by the following variables: 2,3-
81
Fig. 7 – Scatter plot (PC1 and PC2 plotted on x and y axis,
respectively) of PCA scores associated with differentiation
of authentic and non-authentic Brazilian (authentic sample
codes 1–11; non-authentic sample codes 12–13) and
Colombian (authentic sample codes 14–21; non-authentic
sample codes 22–33) coffee samples.
diethylpyrazine, 2-methylpyrazine, 2,3-dimethylpyrazine. PC2
on the other hand was represented as the linear combination
of furfuryl acetate and 5-methylfurfural and it accounted for
33.383% of the variance. The results of the current classifica-
tion study are summarized in the scatter plot illustrated in
Fig. 7.
The results represented in Fig. 7 demonstrate more promi-
nent discrimination of authentic samples from Brazil and
Colombia, which in this case is probably due to the larger num-
ber of samples submitted to this particular PCA evaluation as
compared to the one presented in Section 3.4.1. It is well estab-
lished that the overall success of PCA classification study is
highly correlated with the number of samples submitted to a
particular evaluation.
The outcome scatter plot illustrated in Fig. 7 also demon-
strates that the two non-authentic samples corresponding to
Brazilian origin specification (sample codes 12 and 13) and
collected from Canadian coffee import industries are grouped
very well with the authentic samples from Brazil. However, the
similar finding cannot be confirmed for the Colombian non-
authentic samples (sample codes 22–33) belonging to Canada
import collection. These samples deviate strongly from the
corresponding authentic collection, except for the sample
labelled with code 28, which in this case is closest to the
Colombian authentic collection grouping.
Furthermore, except for the sample 28 which is more asso-
ciated with the authentic sample grouping and outlier sample
22 which is located on the highly positive range of PC2, the
remaining non-authentic Colombian samples are strongly cor-
related with each other. Considering that these samples form
a distinct cluster on the negative ranges of PC1 and PC2, it can
be concluded that their volatile and semivolatile compositions
were potentially altered during transportation or storage con-
ditions. It has been previously documented that both green
and roasted coffee types are prone to compositional changes
82 a n a l y t i c a c h i m i c a a c t a 6 1 7 ( 2 0 0 8 ) 72–84

during storage [70]. The amount of published studies in this

field is scarce for the definite conclusion concerned with the
nature of changes to be drawn [71]. However, it is known that
the storage conditions decrease the roasting coffee aroma
and lead to the production of stale and off-flavour notes. The
extent to which the shelf-life stability of coffee is altered dur-
ing storage is strongly correlated with temperature conditions
as well as moisture and headspace oxygen contents [70]. The
influence of storage on deterioration of true aroma quality
was studied by Reed et al. who reported a gradual decrease
and increase of pyrazine and off-flavour fractions, respectively
during storage of peanuts [72]. The conflicting finding was
presented by Warner et al. who reported constant levels of
pyrazines and increasing levels of aldehydes with storage of
peanuts [73].

3.4.3. Geographical origin assessment of non-authentic

sample collections
Recently, the cultivation of coffee beans has been very promi-
nent in Central America, Africa and Asia. Due to the detection
of increasing practices of selling coffees on the basis of their
geographical origin, the reliability of the current methodol-
ogy for verification of geographical origin traceability was
assessed. For this reason, the non-authentic sample collec-
tions having Central American, African and Asian product
declarations together with non-authentic South American
samples were submitted to statistical evaluation. The objec-
tive of the current examination was the identification of
variables responsible for clustering and discrimination of
samples originating from the following regions: (i) South
America, Colombia (12 samples); (ii) Central America, Costa
Rica and Guatemala (3 and 4 samples, respectively); (iii)
Africa, Ethiopia (3 samples) and (iv) Asia, Indonesia (4 sam-
ples).
The results of the finalized PCA scores are illustrated in
the scatter plot demonstrated in Fig. 8. Two principal compo-
nents were extracted explaining 64.396% of the total variance
of the data set. PC1 represented 35.280% of the variance and
was strongly characterized by 2-ethyl-3,5-dimethylpyrazine
and furfuryl acetate. PC2 explained 29.116% of the variance
and was mainly represented by the linear combination of the
following variables: 2,6-diethylpyrazine, 2-methylpyrazine,
2,3-dimethylpyrazine and 2,5-dimethylpyrazine. The results
represented in scatter plot in Fig. 8 demonstrate a clear classi-
fication of coffee samples according to the geographical origin
attributes. The only outlier that was detected in this partic-
ular study is the sample coded with number 1, which was
already identified in Section 3.4.2 (in Section 3.4.2 this sam-
ple was coded with number 22) as being incorrectly grouped.
The results demonstrated in Fig. 8 illustrate that this sam-
ple for which the Colombian product declaration and origin
specification were claimed by importers is grouped together
with the Ethiopian samples on the highly positive range of
PC1. The grouping of this particular sample which is not estab-
lished in accordance with its geographical origin declaration
demonstrates an example of potential fraudulent practices
that are likely to occur during the distribution of imported food
commodities. The possibility that this sample was mislabelled
during the sample collection procedure might also explain the
incorrect clustering associated with it.
Fig. 8 – Scatter plot (PC2 and PC1 plotted on x and y axis,
respectively) of PCA scores associated with differentiation
of non-authentic South America, Colombia (sample codes
1–12); Central America, Cost Rica and Guatemala (sample
codes 13–15 and 16–19, respectively); Africa, Ethiopia
(sample codes 20–22); and Asia, Indonesia (sample codes
23–26) coffee samples.
In accordance with the previous literature findings, the
discrimination of coffees on the basis of their geographi-
cal origin was established after detecting 5-methylfurfural
among variables responsible for differentiation [11]. In the
current study, the characterization of Ethiopian coffee sam-
ples (sample codes 20–22) and their clustering on the
highly positive side of PC1 was strongly influenced by 2-
ethyl-3,5-dimethylpyrazine and furfuryl acetate. In previously
published literature sources, 2-ethyl-3,5-dimethylpyrazine
was already identified as one of the most potent odorants in
coffee [8], whereas furfuryl acetate was identified as a marker
in coffee origin classification studies [14].
4. Conclusions
Recently, food authenticity has become an issue of signifi-
cant importance as consumers, producers and other parties
within the food sector have become interested in the safety
and provenance of food commodities, thus creating a demand
for implementation of food quality certification procedures.
Furthermore, this issue becomes even more crucial for
import–export trade sectors for which the improvement of
current and generation of novel traceability systems is a must.
Coffee, being one of the most consumable beverages is also
among the food commodities that are subjective to adulterate
and falsifying practices under both compositional and geo-
graphical aspects. Therefore, the objective of the current study
was to characterize coffees on the basis of their geographical
origin attributes and thus verify the product declarations.
For this purpose, HS-SPME–GC–TOFMS methodology was
developed and optimized for the purposes of verifying its
capability in terms of tracing back the coffee samples to their
production area. Among the tested commercially available
 a n a l y t i c a c h i m i c a a c t a 6 1 7 ( 2 0 0 8 ) 72–84
fiber coatings, the mixed-phase DVB/CAR/PDMS fiber demon-
strated best performance characteristics for a wide range
of analytes having different physico-chemical characteris-
tics and hence this coating was used in superelastic metal
fiber assembly form for the completion of overall sequence
of coffee samples. The SPME method optimization was com-
pleted by the utilization of multivariate experimental design
and accordingly the optimum set of conditions for the two
identified influential parameters was 12 min and 55 ◦ C for
extraction time and temperature, respectively. The utiliza-
tion of high-speed data acquisition rate option offered by
Pegasus III TOFMS instrument ensured the completion of
one GC–MS run of a complex coffee sample in 7.9 min. The
complete list of benefits provided by ChromaTOF software
including fully automated background subtraction, baseline
correction, peak find and mass spectral deconvolution algo-
rithms was exploited during the data evaluation procedure.
Finally, the acquired data set was submitted to principal com-
ponent analysis and the corresponding geographical origin
discriminations of coffees originating from South and Cen-
tral America, Africa and Asia were successfully established.
At this point it is important to emphasize that in addition
to successful geographical discrimination of: (i) authentic
sample collections from Brazil and Colombia and (ii) non-
authentic sample collections from South America, Central
America, Africa and Asia, this classification study was also
successful in detecting potential compositional changes that
coffee undergoes due to the limited shelf-life stability over
extensive storage conditions. This finding is crucial in real-
ization that imported food commodities are quite likely to
undergo authentic aroma losses before they are distributed
to consumer populations. Finally, the conducted geographi-
cal origin verification of collected samples proved that this
rapid analytical methodology demonstrates great potential for
the assessment of quality and detection of adulterations in
worldwide coffee industry.
Acknowledgements
The authors acknowledge the assistance of LECO (St. Joseph,
MI, USA), Natural Sciences and Engineering Research Council
of Canada (NSERC) and Leap Technologies (Carrboro, NC, USA).
In addition, the authors would like to thank Dr. Luigi Mondello
(Università degli Studi di Messina, Messina, Italy) for his assis-
tance associated with confirmation of analyte identity using
retention index database.
references
[1] I.S. Arvanitoyannis, O.B. Vaitsi, Crit. Rev. Food Sci. 47 (2007)
675.
[2] A.P. Fernandes, M.C. Santos, S.G. Lemos, M.M.C. Ferreira,
A.R.A. Nogueira, J.A. Nóbrega, Spectrochim. Acta, Part B 60
(2005) 717.
[3] L. Mondello, A. Casilli, P.Q. Tranchida, P. Dugo, R. Costa, S.
Festa, G. Dugo, J. Sep. Sci. 27 (2004) 442.
[4] L. Mondello, R. Costa, P.Q. Tranchida, P. Dugo, M.L. Presti, S.
Festa, A. Fazio, G. Dugo, J. Sep. Sci. 28 (2005) 1101.
[5] I. Esteban-Dı́ez, J.M. González-Sáiz, C. Pizarro, Anal. Chim.
Acta 514 (2004) 57–67.
83
[6] M.J. Martı́n, F. Pablos, A.G. Gonzáles, Characterization of
arabica and robusta roasted coffee varieties and mixture
resolution according to their metal content, Food Chem. 66
(1999) 365.
[7] M.J. Martı́n, F. Pablos, A.G. Gonzáles, Anal. Chim. Acta 358
(1998) 117.
[8] L. Maeztu, C. Sanz, S. Andueza, M. Paz De Peña, J. Bello, C.
Cid, J. Agric. Food Chem. 49 (2001) 5437.
[9] C.G. Zambonin, L. Balest, G.E. De Benedetto, F. Palmisano,
Talanta 66 (2005) 261.
[10] S. Rocha, L. Maetzu, A. Barros, C. Cid, M.A. Coimbra, J. Sci.
Food Agric. 84 (2003) 43.
[11] A.M. Costa Freitas, C. Parreira, L. Vilas-Boas, J. Food Compos.
Anal. 14 (2001) 513.
[12] I. Blank, W. Grosch, J. Agric. Food Chem. 50 (2002) 4653.
[13] S. Casal, M.B.P.P. Oliveira, M.R. Alves, M.A. Ferreira, J. Agric.
Food Chem. 48 (2000) 3420.
[14] A.M. Costa Freitas, A.I. Mosca, Food Res. Int. 32 (1999) 565.
[15] C.P. Bicchi, O.M. Panero, G.M. Pellegrino, A.C. Vanni, J. Agric.
Food Chem. 45 (1997) 4680.
[16] L.M. Reid, C.P. O’Donnell, G. Downey, Trends Food Sci. Tech.
17 (2006) 344.
[17] J. Stoffelsma, G. Sipma, D.K. Kettenes, J. Pypker, J. Agric. Food
Chem. 6 (1968) 1000.
[18] K. Kumazawa, H. Masuda, J. Agric. Food Chem. 51 (2003)
3079.
[19] C. Sanz, D. Ansorena, J. Bello, C. Cid, J. Agric. Food Chem. 49
(2001) 1364.
[20] F. Mayer, W. Grosch, Flavour Fragr. J. 16 (2001) 180.
[21] C. Bicchi, C. Iori, P. Rubiolo, P. Sandra, J. Agric. Food Chem. 50
(2002) 449.
[22] H. Kataoka, H.L. Lord, J. Pawliszyn, J. Chromatogr. A 880
(2000) 35.
[23] F. Bianchi, M. Careri, C. Conti, M. Musci, R. Vreuls, J. Sep. Sci.
30 (2007) 527.
[24] A. Akiyama, K. Murakami, M. Ikeda, K. Iwatsuki, A. Wada, K.
Tokuno, M. Onishi, H. Iwabuchi, J. Food Sci. Tech., Part C 72
(2007) 388.
[25] J. Baggenstoss, L. Poisson, R. Luethi, R. Perren, F. Escher, J.
Agric. Food Chem. 55 (2007) 6685.
[26] M. Charles-Bernard, D.D. Roberts, K. Kraehenbuehl, J. Agric.
Food Chem. 53 (2005) 4426.
[27] O. Gonzalez-Rios, M.L. Suarez-Quiroz, R. Boulanger, M. Barel,
B. Guyot, J. Guiraud, S. Schorr-Galindo, J. Food Compos. Anal.
20 (2007) 289.
[28] J. Meija, J.M. Bryson, A.P. Vonderheide, M. Montes-Bayón, J.A.
Caruso, J. Agric. Food Chem. 51 (2003) 5116.
[29] M. Akiyama, K. Murakami, N. Ohtani, K. Iwatsuki, K.
Sotoyama, A. Wada, K. Tokuno, H. Iwabuchi, K. Tanaka, J.
Agric. Food Chem. 51 (2003) 1961.
[30] I. López-Galilea, N. Fournier, C. Cid, E. Guichard, J. Agric.
Food Chem. 54 (2006) 8560.
[31] M. Akiyama, K. Murakami, M. Ikeda, K. Iwatsuki, S. Kokubo,
A. Wada, K. Tokuno, M. Onishi, H. Iwabuchi, K. Tanaka, Food
Sci. Technol. Res. 11 (2005) 298.
[32] D. Ryan, R. Shellie, P. Tranchida, A. Casilli, L. Mondello, Philip
Marriott, J. Chromatogr. A 1054 (2004) 57.
[33] W. Wardencki, M. Michulec, J. Curylo, Int. J. Food Sci. Tech. 39
(2004) 703.
[34] F. Pellati, S. Benvenuti, F. Yoshizaki, D. Bertelli, M.C. Rossi, J.
Chromatogr. A 1087 (2005) 265.
[35] S.L.C. Ferreira, W.N.L. Dos Santos, C.M. Quintella, B.B. Neto,
J.M. Bosque-Sendra, Talanta 63 (2004) 1061.
[36] E.T. Sousa, F.M. Rodrigues, C.C. Martins, F.S. Oliviera, P.A.P.
Pereira, J.B. Andrade, Microchem. J. 82 (2006) 142.
[37] L. Setkova, S. Risticevic, J. Pawliszyn, J. Chromatogr. A 1147
(2007) 213.
[38] E. Carasek, J. Pawliszyn, J. Agric. Food Chem. 54 (2006) 8688.
84 a n a l y t i c a c h i m i c a a c t a 6 1 7 ( 2 0 0 8 ) 72–84
[39] J. Pawliszyn, Solid Phase Microextraction – Theory and
Practice, Wiley, New York, 1997.
[40] L. Setkova, S. Risticevic, J. Pawliszyn, J. Chromatogr. A 1147
(2007) 224.
[41] F. Bianchi, M. Careri, A. Mangia, M. Musci, J. Sep. Sci. 30
(2007) 563.
[42] R. Shellie, L. Mondello, P. Marriott, G. Dugo, J. Chromatogr. A
970 (2002) 225.
[43] J.A. Pino, R. Marbott, J. Agric. Food Chem. 49 (2001) 5880.
[44] R.J. Bartelt, Anal. Chem. 69 (1997) 364.
[45] C. Pérès, C. Viallon, J. Berdagué, Anal. Chem. 73 (2001) 1030.
[46] D.T. Stanton, P.C. Jurs, Anal. Chem. 61 (1989) 1328.
[47] S. Mihara, N. Enomoto, J. Chromatogr. A 324 (1985) 428.
[48] B. Remberg, A. Nikiforov, G. Buchbauer, Sci. Pharm. 64 (1996)
51.
[49] Y. Tokitomo, M. Steinhaus, A. Büttner, P. Schieberle, Biosci.
Biotechnol. Biochem. 69 (2005) 1323.
[50] P. Schnermann, P. Schieberle, J. Agric. Food Chem. 45 (1997)
867.
[51] R. Triqui, G.A. Reineccius, J. Agric. Food Chem. 43 (1995)
1883.
[52] H.H. Baek, K.R. Cadwallader, E. Marroquuin, J.L. Silva, J. Food
Sci. 62 (1997) 249.
[53] M.J. Jordán, C.A. Margarı́a, P.E. Shaw, K.L. Goodner, J. Agric.
Food Chem. 50 (2002) 5386.
[54] E. Tudor, J. Chromatogr. A 779 (1997) 287.
[55] M.J. Jordán, C.A. Margarı́a, P.E. Shaw, K.L. Goodner, J. Agric.
Food Chem. 51 (2003) 1421.
[56] K. Héberger, M. Görgényi, J. Chromatogr. A 845 (1999) 21.
[57] M. Pompe, J. Chem. Inf. Comput. Sci. 39 (1999) 59.
[58] S. Krist, S. Bail, H. Unterweger, M.B. Ngassoum, A.M.
Mohagir, G. Buchbauer, Eur. J. Lipid Sci. Technol. 108 (2006)
583.
[59] R.V. Golovnya, T.A. Misharina, I.V. Beletskiy,
Chromatographia 34 (1992) 497.
[60] I.L. Zhuravleva, T.E. Kuz’menko, Russ. Chem. B+ 48 (1999)
1931.
[61] A. Radecki, J. Grzybowski, J. Chromatogr. A 152 (1978) 211.
[62] L. Mondello, The FFNSC mass spectra library: an innovative
and reliable peak assignment procedure.
[63] J.A. Pino, R. Marbot, C. Vázquez, J. Agric. Food Chem. 50
(2002) 6023.
[64] C.T. Peng, S.F. Ding, R.L. Hua, Z.C. Yang, J. Chromatogr. A 436
(1988) 137.
[65] W. Jennings, T. Shibamoto, Qualitative Analysis of Flavor and
Fragrance Volatiles by Glass Capillary Gas Chromatography,
Academic Press, New York, 1980.
[66] K. Klesk, M. Qian, R.R. Martin, J. Agric. Food Chem. 52 (2004)
5155.
[67] R. Kowalski, T. Wolski, Flavour Fragr. J. 20 (2005) 306.
[68] R.A. Shellie, P.J. Marriott, Analyst 128 (2003) 879.
[69] Á. Högnadóttir, R.L. Rouseff, J. Chromatogr. A 998 (2003) 201.
[70] R.J. Clarke, Coffee: Handbook of Food and Beverage Stability,
Academic Press, London, 1986.
[71] H.D. Belitiz, W. Grosch, Coffee, Tea, Cocoa: Food Chemistry,
Springer-Verlag, Heidelberg, 1987.
[72] K.A. Reed, C.A. Sims, D.W. Gorbet, S.F. O’Keefe, Food Res. Int.
35 (2002) 769.
[73] K.J.H. Warner, P.S. Dimick, G.R. Ziegler, R.O. Mumma, R.
Hollender, J. Food Sci. 61 (1996) 469.